An intelligent book recommendation system and method based on library reader behavior analysis

Abstract

This invention provides an intelligent book recommendation system and method based on library reader behavior analysis. In the field of intelligent recommendation technology, this invention analyzes the semantic associations of books in historical borrowing transactions and retrieval sessions to extract a semantic transfer pattern dictionary representing knowledge exploration paths. This allows recommendations to move beyond simple similarity matching and guide readers to discover new knowledge in relevant fields, breaking down information silos. Secondly, it can capture and integrate multiple reader behavior patterns in real time, including borrowing, reservation, and collection behaviors, generating real-time interest anchor vectors that enable the recommendation system to dynamically respond to changes in current interests. Furthermore, by matching and overlaying individual interest anchors with group semantic transfer patterns, it generates diffused interest vectors that both match individual current interests and incorporate group knowledge exploration experience. This results in a final recommendation list that not only has personalized features but also includes interdisciplinary and exploratory content, enriching the diversity of recommendations.

Description

Technical Field

[0001] This invention relates to the field of intelligent recommendation technology, and in particular to an intelligent book recommendation system and method based on library reader behavior analysis. Background Technology

[0002] Traditional library book recommendation systems primarily rely on two main technical approaches. The first is content-based filtering, which recommends books by analyzing the similarity between book metadata and readers' borrowing history. While this ensures the relevance of recommendations to users' past interests, it has limitations, namely, a high risk of information cocoon effect, making it difficult to provide novel recommendations beyond readers' existing interests. The second is collaborative filtering, including user-based and item-based collaborative filtering. This method relies on discovering similar user groups or sets of similar items for recommendations, but its effectiveness is hampered by data sparsity and cold start issues, failing to generate effective recommendations for newly added books or new readers. Furthermore, most existing methods are limited to single-point analysis of explicit rating data or borrowing records, failing to delve into the potential interest exploration paths exhibited by readers during their browsing process and the implicit semantic connections between different knowledge domains. This results in recommendations that often remain at a superficial relevance level, lacking an understanding and guidance of readers' deeper knowledge needs. As library collections become increasingly digitized and enriched, readers face a more severe problem of information overload. How to quickly and accurately discover reading materials that meet their real needs and have knowledge expansion value from massive resources has become a key challenge for improving modern smart library services.

[0003] Therefore, it is necessary to provide an intelligent book recommendation system and method based on library reader behavior analysis to solve the above-mentioned technical problems. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides an intelligent book recommendation system and method based on library reader behavior analysis, achieving the beneficial effects of personalized, intelligent book recommendations with knowledge discovery guidance functions.

[0005] This invention provides an intelligent book recommendation method based on library reader behavior analysis, comprising:

[0006] S1: Divide the library's historical borrowing records into real-name borrowing transactions and anonymous search logs into anonymous session transactions. Based on the pre-built semantic vector library of the entire library, query to obtain the semantic content vectors of all books in each transaction, and form the semantic content vector set corresponding to each transaction.

[0007] S2: Calculate the semantic displacement vector between each pair of vectors in the semantic content vector set corresponding to each transaction, and perform weighted clustering analysis on all semantic displacement vectors based on the preset confidence weight to obtain a semantic transfer pattern dictionary composed of multiple cluster centers.

[0008] S3: Capture the real-time behavioral pattern sequence of the target reader's identifiable identity in real time, and based on the preset behavioral pattern weights, weight and fuse the semantic content vectors of all books associated with the real-time behavioral pattern sequence to obtain the real-time interest anchor vector.

[0009] S4: Calculate the first similarity between the real-time interest anchor vector and each semantic transfer pattern in the semantic transfer pattern dictionary, and select the top L semantic transfer patterns arranged in reverse order of the first similarity to form a candidate activation transfer pattern set;

[0010] S5: Weight the real-time interest anchor vector and all activation migration mode vectors in the candidate activation migration mode set to generate a diffusion interest vector;

[0011] S6: Using the approximate nearest neighbor search algorithm, search in the pre-built semantic vector database of the entire library to obtain the top N books with the second similarity to the diffusion interest vector in reverse order, forming a candidate recommendation set. Combine the book popularity and newness to sort the candidate recommendation set and generate the final recommendation list.

[0012] Preferably, in step S1, the steps for dividing real-name borrowing transactions include:

[0013] Each borrowing record in the library's historical borrowing records is treated as an independent real-name borrowing transaction, and each independent real-name borrowing transaction is associated with a real-name authenticated reader's information.

[0014] Preferably, in step S1, the steps for dividing anonymous session transactions include:

[0015] In the anonymous search logs generated on the same terminal, a sequence of search behaviors in which the time interval between any two adjacent search behaviors is less than or equal to a preset continuous session time threshold within a preset time window is aggregated into an anonymous session transaction.

[0016] Preferably, in step S1, the semantic content vector is a dense vector generated by a pre-trained natural language processing model based on the book's title, abstract, and keywords.

[0017] Preferably, in step S2, the preset confidence weights include:

[0018] A first confidence weight is assigned to the semantic displacement vector generated from the real-name borrowing transaction, and a second confidence weight is assigned to the semantic displacement vector generated from the anonymous session transaction, wherein the first confidence weight is greater than the second confidence weight.

[0019] Preferably, in step S3, the real-time behavioral pattern sequence of identifiable identity includes: book borrowing behavior, book reservation behavior, and adding books to a personal book list behavior.

[0020] Preferably, in step S3, the step of obtaining the real-time interest anchor vector includes:

[0021] Real-time monitoring of identifiable behavioral patterns generated by readers using their personal identification credentials within the library system;

[0022] Obtain the books associated with each behavior pattern in the real-time behavior pattern sequence, and query the semantic content vector of each associated book from the pre-built semantic vector library of the entire library;

[0023] Based on preset behavioral pattern weights, the semantic content vectors of all related books are weighted and fused to obtain real-time interest anchor vectors. The preset behavioral pattern weights include borrowing behavior weight, reservation behavior weight, and adding to book list behavior weight.

[0024] Preferably, in step S5, the weighting weight of each activated migration mode vector in the weighted superposition is obtained by normalizing the first similarity between each activated migration mode vector and the real-time interest anchor vector.

[0025] Preferably, in step S6, the step of sorting the candidate recommendation set includes:

[0026] The total number of borrowings and circulation months of each book in the candidate recommendation set are counted. The total number of borrowings of each book is increased by 1, and its natural logarithm is taken. Then, this logarithm is divided by the circulation months of the book to obtain the popularity score of each book.

[0027] For each book in the candidate recommendation set, the book age is obtained by subtracting the book's publication year from the current year. Based on the book age and the preset decay coefficient, the newness score of each book is calculated using an exponential decay function with the natural constant e as the base.

[0028] The second similarity between each book in the candidate recommendation set and the diffusion interest vector is used as the content relevance score of each book. The content relevance score, popularity score and novelty score are weighted and fused to obtain the comprehensive recommendation score of each book.

[0029] The final recommendation list is obtained by sorting each book in the candidate recommendation set in reverse order based on the comprehensive recommendation score.

[0030] This invention provides an intelligent book recommendation system based on library reader behavior analysis, comprising:

[0031] The behavioral transaction module is used to divide the library's historical borrowing records into real-name borrowing transactions and anonymous search logs into anonymous session transactions. Based on a pre-built semantic vector library of the entire library, it queries to obtain the semantic content vectors of all books in each transaction, forming the semantic content vector set corresponding to each transaction.

[0032] The pattern mining module is used to calculate the semantic displacement vector between pairs of vectors in the semantic content vector set corresponding to each transaction, and to perform weighted clustering analysis on all semantic displacement vectors based on preset confidence weights to obtain a semantic transfer pattern dictionary composed of multiple cluster centers.

[0033] The interest anchor module is used to capture real-time behavioral pattern sequences of identifiable target readers. Based on preset behavioral pattern weights, the semantic content vectors of all books associated with the real-time behavioral pattern sequences are weighted and fused to obtain real-time interest anchor vectors.

[0034] The pattern matching module is used to calculate the first similarity between the real-time interest anchor vector and each semantic transfer pattern in the semantic transfer pattern dictionary, and select the top L semantic transfer patterns arranged in reverse order of the first similarity to form a candidate activation transfer pattern set.

[0035] The interest diffusion module is used to weight and superimpose the real-time interest anchor vector with all the activation migration pattern vectors in the candidate activation migration pattern set to generate a diffused interest vector.

[0036] The recommendation generation module uses an approximate nearest neighbor search algorithm to search a pre-built semantic vector database of all books in the library, obtain the top N books with the second similarity to the diffusion interest vector in reverse order, form a candidate recommendation set, and sort the candidate recommendation set by combining book popularity and newness to generate the final recommendation list.

[0037] Compared with related technologies, the intelligent book recommendation system and method based on library reader behavior analysis provided by this invention has the following beneficial effects:

[0038] This invention constructs a recommendation mechanism based on semantic transfer patterns by deeply mining and analyzing reader behavior data. This method effectively integrates real-name borrowing records and anonymous search logs, fully leveraging collective intelligence to infer the potential interests of individual readers. First, by analyzing the semantic associations of books in historical borrowing transactions and search sessions, a semantic transfer pattern dictionary representing knowledge exploration paths is successfully extracted. This allows the recommendation process to move beyond simple similarity matching and guide readers to discover new knowledge in relevant fields, effectively breaking down information silos. Second, it can capture and integrate multiple reader behavior patterns in real time, including borrowing, reservation, and collection behaviors. Through weighted real-time interest anchor vectors, the recommendation system can dynamically respond to changes in current interests, greatly enhancing the timeliness and effectiveness of recommendations. Furthermore, by matching and overlaying individual interest anchors with group semantic transfer patterns, a diffusion interest vector can be generated that not only matches the individual's current interests but also incorporates the group's knowledge exploration experience. This makes the final recommendation list not only highly personalized but also includes interdisciplinary and exploratory content, enriching the diversity of recommendations. Finally, the system uses a multi-factor optimization ranking algorithm to organically combine content relevance, book popularity, and novelty indicators, ensuring that the recommendation results are both accurate and novel, while also balancing classic works with the latest resources. Overall, this invention not only solves the common data sparsity and cold start problems in traditional recommendation systems but also provides readers with more inspiring and guiding knowledge discovery services by introducing semantic analysis and group behavior pattern mining, making recommendations more scientific and accurate. Attached Figure Description

[0039] Figure 1 This is a flowchart of an intelligent book recommendation method based on library reader behavior analysis according to the present invention;

[0040] Figure 2 This is a module structure diagram of an intelligent book recommendation system based on library reader behavior analysis according to the present invention. Detailed Implementation

[0041] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the drawings, not all structures. Moreover, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0042] It should also be noted that, for ease of description, the accompanying drawings show only the parts relevant to the invention and not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of the operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but it may also have additional steps not included in the drawings. The process may correspond to a method, function, procedure, subroutine, subroutine, etc.

[0043] Example 1:

[0044] An intelligent book recommendation method based on library reader behavior analysis, in its specific implementation process, such as... Figure 1 As shown, it illustrates a flowchart of an intelligent book recommendation method based on library reader behavior analysis, including:

[0045] Step S1: Divide the library's historical borrowing records into real-name borrowing transactions and anonymous search logs into anonymous session transactions. Based on the pre-built semantic vector library of the entire library, query to obtain the semantic content vectors of all books in each transaction, and form the semantic content vector set corresponding to each transaction.

[0046] Specifically, in step S1, the steps for dividing real-name borrowing transactions include:

[0047] Each borrowing record in the library's historical borrowing records is treated as an independent real-name borrowing transaction, and each independent real-name borrowing transaction is associated with a real-name authenticated reader's information.

[0048] Specifically, in step S1, the steps for dividing anonymous session transactions include:

[0049] In the anonymous search logs generated on the same terminal, a sequence of search behaviors in which the time interval between any two adjacent search behaviors is less than or equal to a preset continuous session time threshold within a preset time window is aggregated into an anonymous session transaction.

[0050] Specifically, in step S1, the semantic content vector is a dense vector generated by a pre-trained natural language processing model based on the book's title, abstract, and keywords.

[0051] In the specific implementation process, firstly, a semantic vector library for the entire library is constructed. This construction process is completed offline during the system initialization phase. It is based on natural language processing technology to transform the textual metadata of all books in the library's collection into numerical semantic vector representations. For example, the structured metadata of all books is exported from the library's integrated management system. This metadata includes, but is not limited to, the book's ISBN, main title, subtitle, abstract, author, provided keywords, and Chinese Library Classification number. Then, a pre-trained natural language processing model is used, for example, the Sentence-BERT model based on the Transformer architecture, to concatenate the metadata text information of each book into a complete descriptive text. This text is then input into the pre-trained natural language processing model to generate a dense vector of fixed dimensions. This vector can effectively capture the deep semantic information of the books, making books with similar content closer together in the vector space. All generated book semantic vectors are indexed by the book's ISBN and stored using an efficient data structure, for example, through the FAISS index, forming a semantic vector library for the entire library. This library is regularly updated to include the semantic vectors of newly cataloged books. Next, regarding the classification of real-name borrowing transactions, the system extracts each independent borrowing record from historical borrowing records. Each record includes the borrowing time, reader ID number, and book's ISBN. Each such record is treated as a real-name borrowing transaction unit, and each transaction unit is associated with specific real-name authenticated reader information, ensuring data traceability. For the classification of anonymous session transactions, the system collects anonymous log data generated by the library's retrieval system and public query terminals. It aggregates retrieval behavior using a time-window heuristic method, grouping all retrieval queries and book details generated within a preset time window (e.g., a 30-minute window) from the same terminal IP address. Click events are aggregated, with a five-minute time interval threshold for consecutive search behaviors. This means that when the time interval between two adjacent search behaviors does not exceed five minutes, these behaviors are considered as part of the same session, ultimately forming a complete anonymous session transaction. Each transaction contains the ISBNs of all the books searched and clicked by the user during that session. After the transaction is divided, the system uses a pre-built semantic vector library of the entire library to obtain the corresponding semantic content vector by querying the ISBN of each book involved in the transaction. This allows the system to construct a corresponding semantic content vector set for each transaction. These semantic content vector sets will serve as the basic data source for subsequent semantic displacement analysis and pattern mining.

[0052] Step S2: Calculate the semantic displacement vector between each pair of vectors in the semantic content vector set corresponding to each transaction, and perform weighted clustering analysis on all semantic displacement vectors based on the preset confidence weight to obtain a semantic transfer pattern dictionary composed of multiple cluster centers.

[0053] Specifically, in step S2, the preset confidence weights include:

[0054] A first confidence weight is assigned to the semantic displacement vector generated from the real-name borrowing transaction, and a second confidence weight is assigned to the semantic displacement vector generated from the anonymous session transaction, wherein the first confidence weight is greater than the second confidence weight.

[0055] In the specific implementation process, firstly, it is necessary to calculate the semantic displacement vectors between all pairs of vectors in the semantic content vector set corresponding to each transaction. For the semantic content vector set of each transaction, a complete traversal of combinatorial mathematics is performed, that is, calculating all combinations of semantic content vectors. For each pair of semantic content vectors, the system calculates two semantic displacement vectors in opposite directions. These two semantic displacement vectors physically represent two distinct but equally existing directions of knowledge transfer. This bidirectional calculation ensures that any possible knowledge association paths between all books in the transaction can be completely captured without losing any potential exploration patterns. This semantic displacement vector quantifies the direction and distance from the source book to the target book in the semantic space, representing the transfer path of a reader from one type of knowledge content to another in a transaction. After obtaining the semantic displacement vectors generated by all transactions, weighted clustering analysis is performed. Semantic displacement vectors generated from real-name borrowing transactions are assigned a higher first confidence weight (for example, the first confidence weight is set to 1.0), while semantic displacement vectors generated from anonymous session transactions are assigned a lower second confidence weight (for example, set to 0.3). The difference in weights reflects the different data. The reliability of the source is crucial. Real-name borrowing behavior represents the reader's final knowledge acquisition decision and therefore has higher confidence, while anonymous retrieval behavior represents more exploratory intent and thus has lower confidence. For example, weighted clustering analysis uses an improved weighted K-means clustering algorithm. First, all semantic displacement vectors are preprocessed, associating each semantic displacement vector with its corresponding confidence weight. During clustering, the algorithm fully considers the weight value of each semantic displacement vector. When calculating cluster centers, semantic displacement vectors with higher weights have a greater impact on the location of the cluster center. The number of clusters K is determined through methods including but not limited to the elbow rule and silhouette coefficient analysis to ensure that the mined patterns are neither too general nor too fragmented. After clustering, each cluster center represents a typical semantic transfer pattern. These patterns reveal frequently occurring knowledge exploration paths within the reader group. Finally, all cluster center vectors are organized into a semantic transfer pattern dictionary, providing the knowledge foundation for subsequent personalized recommendations. The entire weighted clustering process ensures that the final semantic transfer patterns focus more on those more representative knowledge transfer paths generated by genuine borrowing behavior, thereby improving the accuracy and reliability of the recommendation system.

[0056] Step S3: Capture the real-time behavioral pattern sequence of the target reader's identifiable identity in real time. Based on the preset behavioral pattern weights, weight and fuse the semantic content vectors of all books associated with the real-time behavioral pattern sequence to obtain the real-time interest anchor vector.

[0057] Specifically, in step S3, the real-time behavioral pattern sequence that can be identified includes: book borrowing behavior, book reservation behavior, and adding books to a personal book list behavior.

[0058] Specifically, in step S3, the steps for obtaining the real-time interest anchor vector include:

[0059] Real-time monitoring of identifiable behavioral patterns generated by readers using their personal identification credentials within the library system;

[0060] Obtain the books associated with each behavior pattern in the real-time behavior pattern sequence, and query the semantic content vector of each associated book from the pre-built semantic vector library of the entire library;

[0061] Based on preset behavioral pattern weights, the semantic content vectors of all related books are weighted and fused to obtain real-time interest anchor vectors. The preset behavioral pattern weights include borrowing behavior weight, reservation behavior weight, and adding to book list behavior weight.

[0062] In the specific implementation process, firstly, the library's identity authentication system captures the real-time behavioral pattern sequence of the target reader's current identifiable identity. These behaviors include book borrowing, book reservation, and adding books to a personal book list. The system continuously monitors these operations while the reader is logged in, forming a real-time behavioral stream arranged in chronological order. Next, for each behavioral pattern in the sequence, the directly associated book identifier is extracted. For example, the ISBN of books in the borrowing record and the book entries in the reserved book list are extracted, and the pre-built semantic vector library of the entire library is immediately queried to obtain the semantic content vectors corresponding to these books. Then, a weighted fusion is performed based on the preset behavioral pattern weights, where book borrowing has the highest weight, representing the reader's final knowledge acquisition decision; book reservation has the second highest weight, reflecting the reader's strong intention; and adding books to a book list has a lower weight, representing potential interest. The weighted fusion adopts a weighted average method, that is, each semantic content vector is multiplied by the weight of its corresponding behavior, summed, and then divided by the sum of all weights, thereby generating a comprehensive real-time interest anchor vector, which accurately encodes the reader's current comprehensive interest tendency.

[0063] Step S4: Calculate the first similarity between the real-time interest anchor vector and each semantic transfer pattern in the semantic transfer pattern dictionary, and select the top L semantic transfer patterns arranged in reverse order of the first similarity to form a candidate activation transfer pattern set.

[0064] In the specific implementation process, firstly, each semantic transfer pattern vector in the semantic transfer pattern dictionary is traversed, and the first similarity between each vector and the real-time interest anchor vector is calculated. For example, the first similarity is calculated using the cosine similarity algorithm, which evaluates the directional consistency by measuring the cosine value of the angle between the two vectors in space, and finally obtains a similarity score between -1 and 1. The closer the similarity score is to 1, the more consistent the semantic directions of the two vectors are. After completing the calculation of all semantic transfer patterns and real-time interest anchor vectors in the semantic transfer pattern dictionary, a list containing each semantic transfer pattern and its corresponding similarity score is obtained. Then, the list of similarity scores is sorted in descending order according to the similarity score, so that the pattern most relevant to the real-time interest is placed at the front of the list. After sorting, the top L semantic transfer patterns are selected according to the preset threshold strategy to construct a candidate activation transfer pattern set, where the value of L is determined by a pre-set configuration, for example, set to a fixed integer of 10. The obtained candidate activation transfer pattern set provides the data foundation for subsequent steps.

[0065] Step S5: Weight and superimpose the real-time interest anchor vector with all the activation migration mode vectors in the candidate activation migration mode set to generate a diffusion interest vector.

[0066] Specifically, in step S5, the weighting weight of each activated migration mode vector in the weighted superposition is obtained by normalizing the first similarity between each activated migration mode vector and the real-time interest anchor vector.

[0067] In the specific implementation process, firstly, the first similarity value between the real-time interest anchor vector and each activation migration pattern vector in the candidate activation migration pattern set is obtained from the results of step S4. These similarity values ​​reflect the correlation strength between each migration pattern and the reader's current interest. Subsequently, these first similarity values ​​are normalized to convert them into weight coefficients. For example, the normalization is implemented using the Softmax function, that is, the exponent of each similarity value is calculated and then divided by the sum of the exponents of all similarity values, thereby ensuring that the sum of all weights is 1 and each weight value is between 0 and 1. This allows patterns with high similarity to receive greater weights and patterns with low similarity to receive smaller weights. Next, a weighted summation operation is performed, linearly combining the real-time interest anchor vector with each activation transfer mode vector according to their corresponding weights. The summation iterates through all modes in the candidate activation transfer mode set, and the weighted summation operation shifts the reader's original interest points along the directions of multiple related transfer modes, thereby generating a diffused interest vector located at a new position in the semantic space. This diffused interest vector retains the core features of the original interest while incorporating the guiding information of the group's exploration path. The final generated diffused interest vector will serve as the basis for subsequent retrieval, encoding richer and more diverse interest possibilities and providing a broader exploration space for the recommendation system.

[0068] Step S6: Using the approximate nearest neighbor search algorithm, search the pre-built semantic vector database of the entire library to obtain the top N books with the second similarity to the diffusion interest vector in reverse order, forming a candidate recommendation set. Sort the candidate recommendation set by combining book popularity and newness to generate the final recommendation list.

[0069] Specifically, in step S6, the step of sorting the candidate recommendation set includes:

[0070] The total number of borrowings and circulation months of each book in the candidate recommendation set are counted. The total number of borrowings of each book is increased by 1, and its natural logarithm is taken. Then, this logarithm is divided by the circulation months of the book to obtain the popularity score of each book.

[0071] For each book in the candidate recommendation set, the book age is obtained by subtracting the book's publication year from the current year. Based on the book age and the preset decay coefficient, the newness score of each book is calculated using an exponential decay function with the natural constant e as the base.

[0072] The second similarity between each book in the candidate recommendation set and the diffusion interest vector is used as the content relevance score of each book. The content relevance score, popularity score and novelty score are weighted and fused to obtain the comprehensive recommendation score of each book.

[0073] The final recommendation list is obtained by sorting each book in the candidate recommendation set in reverse order based on the comprehensive recommendation score.

[0074] In the specific implementation process, firstly, an approximate nearest neighbor search algorithm is used to search the pre-built semantic vector database of the entire library. This search process is implemented through a pre-established HNSW index structure to quickly find the top M books with the highest cosine similarity to the diffusion interest vector. Here, M is greater than the final required N value to reserve screening space. Then, a multi-dimensional score is calculated for each book in the candidate recommendation set. First, the popularity score is calculated by adding 1 to the total number of historical borrowings of each book, taking the natural logarithm, and then dividing by the number of months the book has been in circulation. The number of months in circulation is calculated from the date the book was acquired to the current date in months. This effectively eliminates the influence of differences in book circulation time on the popularity assessment. Secondly, the newness score is calculated by subtracting the book's publication year from the current year to obtain the book's age, and then applying an exponential decay function. The newness score equals the natural logarithm. The negative preset decay coefficient of number e is multiplied by the power of book age, where, for example, the preset decay coefficient ranges from 0.1 to 0.3, so that newly published books receive higher scores. Finally, the second similarity between each book and the diffusion interest vector is used as the content relevance score. Then, the content relevance score, popularity score, and novelty score are weighted and summed according to preset score weight coefficients to obtain the comprehensive recommendation score for each book. The preset score weight coefficients are adjusted according to the library's recommendation strategy. For example, the content relevance weight is set to the highest, followed by popularity, and then novelty to the lowest. Finally, the top N books are selected according to the comprehensive recommendation score from high to low to generate the final recommendation list, which is pushed to the reader's terminal to ensure that the recommendation results have relevance, popularity, and novelty at the same time, so as to achieve the purpose of intelligent book recommendation based on library reader behavior analysis.

[0075] The working principle of the intelligent book recommendation method based on library reader behavior analysis provided by this invention is as follows:

[0076] First, transaction data is constructed by processing historical borrowing records and anonymous retrieval logs. Each transaction contains semantic content vectors of multiple books, generated by a pre-trained model based on book title summaries and keywords. Then, semantic displacement vectors between book vectors within a transaction are calculated to capture knowledge transfer paths. These displacement vectors are then analyzed using weighted clustering to form a semantic transfer pattern dictionary. Displacement vectors from registered borrowing transactions are assigned higher weights, while those from anonymous sessions are assigned lower weights to reflect data reliability. When a target reader engages in real-time behavior, such as borrowing, reserving, or adding to a book list, the system monitors these behaviors and obtains the semantic content vectors of the relevant books. These vectors are then weighted and fused according to the behavior type to generate a semantic transfer pattern dictionary. A real-time interest anchor vector is generated, and then the similarity between the real-time interest anchor vector and each pattern in the dictionary is calculated. The top L most similar patterns are selected as the candidate activation transfer pattern set. Then, the real-time interest anchor vector and the candidate activation transfer pattern vector are weighted and superimposed based on similarity weights to generate a diffusion interest vector, thereby expanding the scope of interests. Finally, the book library is searched for books similar to the diffusion vector using approximate nearest neighbor search to form a candidate recommendation set. The popularity score, novelty score, and content relevance score of each book are calculated. These three types of scores are weighted, merged, and sorted to generate the final recommendation list, realizing intelligent book recommendation based on library reader behavior analysis that takes into account relevance, popularity, and novelty.

[0077] Example 2:

[0078] An intelligent book recommendation system based on library reader behavior analysis, in its specific implementation, such as... Figure 2 As shown, it illustrates the modular structure of an intelligent book recommendation system based on library reader behavior analysis, including:

[0079] The Behavior Transaction Module 100 is used to divide the library's historical borrowing records into real-name borrowing transactions and anonymous search logs into anonymous session transactions. Based on the pre-built semantic vector library of the entire library, it queries and obtains the semantic content vectors of all books in each transaction, forming the semantic content vector set corresponding to each transaction.

[0080] The pattern mining module 200 is used to calculate the semantic displacement vector between pairs of vectors in the semantic content vector set corresponding to each transaction, and to perform weighted clustering analysis on all semantic displacement vectors based on preset confidence weights to obtain a semantic transfer pattern dictionary composed of multiple cluster centers.

[0081] The interest anchor module 300 is used to capture the real-time behavioral pattern sequence of the target reader's identifiable identity in real time. Based on the preset behavioral pattern weights, the semantic content vectors of all books associated with the real-time behavioral pattern sequence are weighted and fused to obtain the real-time interest anchor vector.

[0082] The pattern matching module 400 is used to calculate the first similarity between the real-time interest anchor vector and each semantic transfer pattern in the semantic transfer pattern dictionary, and select the top L semantic transfer patterns arranged in reverse order of the first similarity to form a candidate activation transfer pattern set.

[0083] The interest diffusion module 500 is used to weight and superimpose the real-time interest anchor vector with all the activation migration mode vectors in the candidate activation migration mode set to generate a diffused interest vector.

[0084] The recommendation generation module 600 is used to retrieve the top N books in the pre-built semantic vector database of the entire library using an approximate nearest neighbor search algorithm. These books are arranged in reverse order of the second similarity with the diffusion interest vector, forming a candidate recommendation set. The candidate recommendation set is then sorted by book popularity and newness to generate the final recommendation list.

[0085] The working principle of the intelligent book recommendation system based on library reader behavior analysis provided by this invention is as follows:

[0086] First, the behavioral transaction module 100 processes the library's historical borrowing records and anonymous search logs, dividing the borrowing records into independent borrowing transactions associated with real-name reader information. Simultaneously, based on time windows and continuous session thresholds, anonymous search logs are aggregated into anonymous session transactions. Then, based on a pre-built library-wide semantic vector database, the semantic content vectors of all books in each transaction are queried, generating a semantic content vector set. Next, the pattern mining module 200 calculates the semantic displacement vectors between pairwise book vectors in each semantic content vector set. These displacement vectors represent the direction and distance of knowledge transfer, and are clustered using weighted clustering analysis. After clustering, a semantic transfer pattern dictionary composed of multiple cluster centers is generated, capturing the knowledge exploration patterns of the group of readers. When a target reader engages in real-time behavior, the interest anchor module 300 monitors their identifiable behavioral sequences, such as borrowing, reservations, or adding to a book list. The system acquires semantic content vectors of relevant books and performs weighted fusion based on behavior type weights to generate real-time interest anchor vectors, accurately representing the reader's current interest pattern. The pattern matching module 400 then calculates the similarity between this anchor vector and each pattern in the semantic transfer pattern dictionary, selecting the top L patterns with the highest similarity to construct a candidate activation transfer pattern set. The interest diffusion module 500 then weights and superimposes these activation pattern vectors with the real-time interest anchor vectors to generate diffusion interest vectors, thereby expanding the scope of interests. Finally, the recommendation generation module 600 uses an approximate nearest neighbor search algorithm to retrieve the top N books similar to the diffusion vectors in the entire library's semantic vector database to form a candidate recommendation set. It then comprehensively calculates a weighted fusion of three types of scores for each book: content relevance score, popularity score, and novelty score. Based on the comprehensive recommendation score, it sorts the books to generate a final recommendation list and pushes it to the corresponding readers.

[0087] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, are implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0088] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0089] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. An intelligent book recommendation method based on library reader behavior analysis, characterized in that, The intelligent book recommendation method comprises the following steps: S1: The historical borrowing records of the library are divided into real-name borrowing transactions, and the anonymous search logs are divided into anonymous session transactions. Based on a pre-constructed library-wide book semantic vector library, the semantic content vectors of all books in each transaction are queried to obtain a semantic content vector set corresponding to each transaction; S2: Calculate the semantic displacement vectors between each transaction corresponding semantic content vector set two vectors, and based on the preset confidence weight, all semantic displacement vectors are weighted and clustered to obtain a semantic migration mode dictionary composed of multiple cluster centers; S3: Real-time capture of the real-time behavior pattern sequence of the identifiable identity of the target reader, based on the preset behavior pattern weight, the semantic content vectors of all books associated with the real-time behavior pattern sequence are weighted and fused to obtain a real-time interest anchor point vector; S4: Calculate the first similarity between the real-time interest anchor point vector and each semantic migration mode in the semantic migration mode dictionary, and select the top L semantic migration modes in the first similarity descending order to form a candidate active migration mode set; S5: The real-time interest anchor point vector and all active migration mode vectors in the candidate active migration mode set are weighted and superimposed to generate a diffusion interest vector; S6: Using the approximate nearest neighbor search algorithm, search in the pre-constructed library-wide book semantic vector library to obtain the top N books in the second similarity descending order of the diffusion interest vector to form a candidate recommendation set, and sort the candidate recommendation set according to the book popularity and newness to generate a final recommendation list. 2.The intelligent book recommendation method based on library reader behavior analysis of claim 1, wherein, In step S1, the division step of the real-name borrowing transaction comprises: Each borrowing record in the historical borrowing records of the library is taken as an independent real-name borrowing transaction, and each independent real-name borrowing transaction is associated with a real-name authenticated reader information. 3.The intelligent book recommendation method based on library reader behavior analysis of claim 2, wherein, In step S1, the division step of the anonymous session transaction comprises: In the anonymous search logs generated on the same terminal, within a preset time window, the time interval between any two adjacent search behaviors is less than or equal to the preset continuous session time threshold, and the search behavior sequence is aggregated into an anonymous session transaction. 4.The intelligent book recommendation method based on library reader behavior analysis of claim 3, wherein, In step S1, the semantic content vector is a dense vector generated by a pre-trained natural language processing model based on the title, abstract, and subject of the book. 5.The intelligent book recommendation method based on library reader behavior analysis of claim 4, wherein, In step S2, the preset confidence weight comprises: The semantic displacement vectors generated from the real-name borrowing transactions are assigned a first confidence weight, and the semantic displacement vectors generated from the anonymous session transactions are assigned a second confidence weight, and the first confidence weight is greater than the second confidence weight. 6.The intelligent book recommendation method based on library reader behavior analysis of claim 5, wherein, In step S3, the real-time behavior pattern sequence of the identifiable identity comprises: book borrowing behavior, book reservation behavior, and adding books to a personal book list. 7.The intelligent book recommendation method based on library reader behavior analysis of claim 6, wherein, In step S3, the obtaining step of the real-time interest anchor point vector comprises: Real-time monitoring of the identifiable real-time behavior pattern sequence of the reader generated in the library system through a personal identity credential; acquire the books associated with each behavior pattern in the real-time behavior pattern sequence, and query the semantic content vectors of each associated book from the pre-constructed library-wide book semantic vector library; weight and fuse the semantic content vectors of all associated books based on preset behavior pattern weights to obtain a real-time interest anchor vector, wherein the preset behavior pattern weights include a borrowing behavior weight, a reservation behavior weight, and a booklist joining behavior weight. 8.The intelligent book recommendation method based on library reader behavior analysis of claim 7, wherein, In step S5, the weighting weight of each activated migration pattern vector in the weighted superposition is a first similarity between each activated migration pattern vector and the real-time interest anchor vector after normalization processing. 9.The intelligent book recommendation method based on library reader behavior analysis of claim 8, wherein, In step S6, the step of sorting the candidate recommendation set includes: statistically acquire the total number of borrowings and the circulation months of each book in the candidate recommendation set, add a value of 1 to the total number of borrowings of each book, take the natural logarithm value, then divide the logarithm value by the circulation months of the book to obtain a heat score of each book; for each book in the candidate recommendation set, subtract the publication year of the book from the current year to obtain the book age of each book, and based on the book age and a preset decay coefficient, calculate a newness score of each book by an exponential decay function with a base of the natural constant e; take the second similarity between each book in the candidate recommendation set and the diffusion interest vector as a content relevance score of each book, and weight and fuse the content relevance score, the heat score, and the newness score to obtain a comprehensive recommendation score of each book; sort each book in the candidate recommendation set in a descending order based on the comprehensive recommendation score to obtain a final recommendation list.

10. An intelligent book recommendation system based on library reader behavior analysis, characterized in that, An intelligent book recommendation system based on the intelligent book recommendation method of any one of claims 1 to 9, the intelligent book recommendation system comprising: a behavior transaction module configured to divide historical borrowing records of the library into real-name borrowing transactions, divide anonymous search logs into anonymous session transactions, query and acquire semantic content vectors of all books in each transaction based on a pre-constructed library-wide book semantic vector library, and construct a semantic content vector set corresponding to each transaction; a pattern mining module configured to calculate semantic displacement vectors between two vectors in the semantic content vector set corresponding to each transaction, and perform weighted clustering analysis on all semantic displacement vectors based on a preset confidence weight to obtain a semantic migration pattern dictionary composed of multiple cluster centers; an interest anchor module configured to capture a real-time behavior pattern sequence of a recognizable identity of a target reader in real time, weight and fuse semantic content vectors of all books associated with the real-time behavior pattern sequence based on preset behavior pattern weights to obtain a real-time interest anchor vector; a pattern matching module configured to calculate a first similarity between the real-time interest anchor vector and each semantic migration pattern in the semantic migration pattern dictionary, and select the first L semantic migration patterns arranged in a descending order of the first similarity to construct a candidate activated migration pattern set; an interest diffusion module configured to perform weighted superposition on the real-time interest anchor vector and all activated migration pattern vectors in the candidate activated migration pattern set to generate a diffusion interest vector. An intelligent book recommendation system based on the intelligent book recommendation method of any one of claims 1 to 9, the intelligent book recommendation system comprising: a behavior transaction module configured to divide historical borrowing records of the library into real-name borrowing transactions, divide anonymous search logs into anonymous session transactions, query and acquire semantic content vectors of all books in each transaction based on a pre-constructed library-wide book semantic vector library, and construct a semantic content vector set corresponding to each transaction; a pattern mining module configured to calculate semantic displacement vectors between two vectors in the semantic content vector set corresponding to each transaction, and perform weighted clustering analysis on all semantic displacement vectors based on a preset confidence weight to obtain a semantic migration pattern dictionary composed of multiple cluster centers; an interest anchor module configured to capture a real-time behavior pattern sequence of a recognizable identity of a target reader in real time, weight and fuse semantic content vectors of all books associated with the real-time behavior pattern sequence based on preset behavior pattern weights to obtain a real-time interest anchor vector; a pattern matching module configured to calculate a first similarity between the real-time interest anchor vector and each semantic migration pattern in the semantic migration pattern dictionary, and select the first L semantic migration patterns arranged in a descending order of the first similarity to construct a candidate activated migration pattern set; an interest diffusion module configured to perform weighted superposition on the real-time interest anchor vector and all activated migration pattern vectors in the candidate activated migration pattern set to generate a diffusion interest vector. The recommendation generation module is configured to utilize an approximate nearest neighbor search algorithm to search in a pre-constructed library-wide book semantic vector library to obtain the top N books in a second similarity descending order of the diffusion interest vector, to form a candidate recommendation set, and to sort the candidate recommendation set in combination with book popularity and newness to generate a final recommendation list.

Images